LLMCov: A Methodology for LLM-Aided DfT Coverage Improvement in Open-Source Designs

Scan insertion is one of the important steps in the development of test infrastructure in modern designs. During the scan insertion step, it is very crucial to resolve all design rule violations as these violations may lead to significant drop in the test coverage. Because of nature of these violations, it becomes a tedious debugging exercise and requires lot of manual expertise. In this work, we propose AI-based solution for seeking solution for these design rule checks. The power of AI-based techniques (such as CNN-based models or LLM-based methods) is getting widely recognized in the area of EDA tools performance enhancement. Therefore, we develop a LLM-aided DRC debugging methodology to aid the scan synthesis tool that in turn assists the ATPG tool to obtain enhanced test coverage values. Experimental results indicate significant improvement in the test coverage with fixes obtained with assistance of LLM.